How Blacklisted Chinese Giants are Accessing Advanced AI Models from OpenAI and Google
In Focus
- OpenAI and Google sell AI services to Chinese subsidiaries in Singapore
- Singapore serves as a neutral AI hub in the Asian region
- Current U.S. rules allow Chinese firms abroad to access frontier AI models
OpenAI and Google have reportedly been selling advanced AI models to Chinese subsidiaries in Singapore. The tech companies confirmed providing AI service to Alibaba, Baidu, and Tencent subsidiaries in Singapore despite their parent firms being blacklisted in the U.S. by the Pentagon.
Why Did OpenAI Provide AI Models to Chinese Companies?
Singapore serves as a neutral AI hub and Chinese subsidiaries that are incorporated in the city state are considered Singaporean firms. The subsidiaries operate under Singaporean law, they pay taxes in the country, and they can enter agreements that their parent firms in mainland China cannot.
This legal standing allows subsidiaries of major Chinese firms to access frontier AI models from American companies. Since U.S. export restrictions do not prohibit Chinese firms from accessing frontier AI models outside mainland China, the sale of AI models to the Chinese subsidiaries remains legal.
News about access to OpenAI AI models by Chinese companies has revived calls in Washington for export controls on frontier AI software.
What Investments Have Google and OpenAI Made in Singapore?
Both Google and OpenAI have made strategic investments in Singapore in recent years. In 2024, OpenAI announced plans to open a Singapore office as part of its global expansion strategy.
The AI firm also committed over $234 million to establish its first applied AI lab outside the United States. As the AI firm provides Chinese companies in Singapore access to its frontier models, it has tightened controls on how those technologies are used elsewhere. Last month, OpenAI suspended API access for Alibaba-affiliated users over concerns they were using model outputs for AI distillation, which is a technique that leverages advanced models to improve rival systems.
OpenAI, which launched GPT-5.6 Sol on July 9, said the suspected activity was reported to the U.S. government. OpenAI also said it prohibits direct access to its models from China. However, the company confirmed that it permits some Chinese-owned companies to use its services in markets where it can enforce its security safeguards.
Google DeepMind opened a regional research hub in Singapore this year. The tech giant said AI services are accessible to companies in Singapore and Hong Kong in line with its usage policies. However, the company said location-based restrictions are not sufficient to stop determined users from evading safeguards.
Unlike Google and OpenAI, Anthropic has been more strict in prohibiting Chinese firms and their foreign subsidiaries from accessing its frontier models. Last month, the U.S. government ordered the company to restrict access to Mythos and Fable 5 AI models by all foreign nationals. The curbs have since been lifted.
What Chinese Access to Frontier Models for U.S. Curbs
Chinese access to frontier AI models through overseas subsidiaries highlights a challenge for U.S. technology controls. While current rules restrict exports of advanced chips, they leave gaps around AI software access in third countries. The issue has intensified calls in Washington for broader export controls on frontier AI models to prevent potential technology transfer to Chinese competitors.
